基于支持向量机的住院精神分裂症患者暴力风险预测模型

Predictive modle for violence risk in hospitalized schizophrenia patients based on support vector machine

中文关键词:  LASSO回归  精神分裂症  暴力风险  预测模型

英文关键词:LASSO regression  Schizophrenia  Violent risk  Prediction model

中文摘要:

      背景 精神分裂症患者的暴力攻击行为通常具有突发性强、难以预测、危害性大及防范难度高等特点，早期识别与准确评估其暴力风险具有重要的临床意义。目的 构建精神分裂症患者暴力风险预测模型，明确影响患者暴力行为发生的关键因素，为临床精准量化评估和早期干预提供参考。方法 纳入2022年3月—2024年9月在太原市精神病医院住院、符合《国际疾病分类（第11版）》（ICD-11）诊断标准的200例精神分裂症患者，构成建模队列。按照7∶3将患者分为训练集（n=140）和测试集（n=60）。基于最小绝对收缩和选择算子（LASSO）回归算法对特征变量进行筛选和降维处理，选取机器学习的支持向量机（SVM）进行模型训练与预测，运用受试者工作特征（ROC）曲线下面积（AUC）、准确度、精确度、灵敏度、特异度、F1值和Brier值评价模型的判别效能。结果 LASSO回归筛选出16个特征变量。Pearson相关分析结果显示，既往暴力行为频次与临床精神症状评分呈正相关（r=0.580，P<0.01），住院治疗依从性与目前病情呈正相关（r=0.550，P=0.003），受教育程度与家庭人均月收入呈正相关（r=0.367，P<0.01）。SVM模型的AUC为0.853，准确度为0.800，精确度为0.810，灵敏度为0.895，特异度为0.636，F1值为0.850，Brier值为0.168。结论 SVM模型在住院精神分裂症患者暴力风险评估中的适用水平和整体预测性能较好，有助于此类患者暴力风险的早期识别。

英文摘要:

      Background The violent aggressive behaviors of patients with schizophrenia usually have the characteristics of suddenness， unpredictability， high severity， and great difficulty in prevention. Early identification and accurate assessment of their risk of violent aggression have significant clinical significance.Objective To construct a predictive model for the violence risk in hospitalized patients with schizophrenia， to identify the key factors influencing the occurrence of violent behavior in these patients， so as to provide references for clinical precise quantitative assessment and early intervention.Methods A total of 200 patients with schizophrenia who were hospitalized at Taiyuan Psychiatric Hospital from March 2022 to September 2024 and met the diagnostic criteria of the International Classification of Diseases， eleventh edition （ICD-11） were collected to form the modeling cohort. They were randomly divided into a training set （n=140） and a test set （n=60） at a ratio of 7∶3. Based on the least absolute shrinkage and selection operator （LASSO） regression algorithm， the feature variables were screened and dimension-reduced. The support vector machine （SVM） from machine learning was selected for model training and prediction. The discrimination efficacy of the model was evaluated by the area under the receiver operating characteristic （ROC） curve （AUC）， accuracy， precision， sensitivity， specificity， F1 value， and Brier value.Results LASSO regression screening identified 16 feature variables. Pearson correlation analysis revealed a positive correlation between prior violent behavior frequency and clinical psychiatric symptom scores （r=0.580， P<0.01）， a positive correlation between hospitalization compliance and current disease status （r=0.550， P=0.003）， and a positive correlation between educational level and family per capita monthly income （r=0.367， P<0.01）. The SVM model achieved an AUC of 0.853， accuracy of 0.800， precision of 0.810， sensitivity of 0.895， specificity of 0.636， F1 value of 0.850， and Brier value of 0.168.Conclusion The SVM model has a relatively high level of applicability and overall predictive performance in the assessment of violent risk in schizophrenia patients， which is helpful for the early identification of violent risks in such patients. ［Funded by Specialized Research Project for Enhancing the Competence of Health Professionals in Taiyuan City （number， Y2023006）］